Re: Regression in longterm 4.19: f2fs: use generic EFSBADCRC/EFSCORRUPTED

[Chao Yu]

On 2019/10/14 5:44, Pavel Machek wrote:
> On Sat 2019-10-12 21:55:24, Andrew Macks wrote:
>> Sorry for version typo in the previous message.
>>
>> In addition to 4.19, the issue was also backported to 4.14 and 5.2.
>>
>> 4.14, 4.19 and 5.2 are all missing the EINVAL fix from 5.3.
> 
> Ouch.
> 
> Well, when I seen the patch, I thought "looks like the bug is not
> serious enough for -stable". I guess I should have spoken up.
> 
> Anyway, I guess we need to either revert  59a5cea41dd0a or backport
> 38fb6d0ea34299d97b too....
> 
> So I guess Greg and lists need to be cc-ed... and 
> 
> Thanks for the report and sorry for the trouble....

I'm so sorry to introduce original bug, the fixing patch ("f2fs: use EINVAL for
superblock with invalid magic") should be backported to stable kernel as soon as
possible.

Thanks,

> 
> 								Pavel
> 
> 
>> Andrew.
>>
>> On Sat, 12 Oct 2019 at 21:39, Andrew Macks <andypoo@xxxxxxxxx> wrote:
>>
>>> Hi - there is a nasty regression which was recently introduced into
>>> longterm 4.19.76.
>>>
>>> 59a5cea41dd0ae706ab83f8ecd64199aadefb493 was committed to 4.19, however it
>>> introduces a regression that filesystems no longer mount if do_mounts
>>> iterates through them after F2FS.  This surfaced on one of my servers as
>>> F2FS superblock check happens before btrfs mount is attempted.
>>>
>>> With this code, my server panicked after kernel upgrade as btrfs mount
>>> wasn't attempted.
>>>
>>> This issue has already been fixed in 5.3 with this patch in July, but it
>>> was missed from the 4.19 backport.
>>>
>>> 38fb6d0ea34299d97b031ed64fe994158b6f8eb3
>>> f2fs: use EINVAL for superblock with invalid magic
>>>
>>> Andypoo.
>>>
>